feat: ossie-sql — Calcite engine that executes OSI queries against any JDBC warehouse

[buggtb]

Restructures the repo into a parent + two child modules and adds
ossie-sql — a Calcite adapter + shelf-state query engine that
executes queries authored against an OSI semantic model against any
JDBC warehouse.

What's new

ossie-sql module

Public API (bi.saiku.ossie.sql):

• OssieEngine — top-level, builder-pattern entry point. Takes
  an OSI YAML (path, Reader, inline string, or already-parsed
  OssieDocument) + a JDBC URL + credentials for the warehouse the
  Ossie datasets are backed by. AutoCloseable.
• OssieQuery — immutable shelf-state request. Builder methods:
  .model() / .factDataset() / .rows() / .columns() /
  .values() (with optional aggregation override) / .filter() /
  .sortByMetric() / .sortByField() / .limit().
• OssieResult — records + column descriptors + generated SQL
  • runtime ms.

Internal (bi.saiku.ossie.sql.internal):

• OssieSchemaFactory — Calcite SchemaFactory that registers
  Ossie datasets as virtual tables. Now accepts either an
  ossieYaml file path or an ossieDocumentInline YAML string
  operand.
• OssieSchema, OssieDatasetTable, OssieMetricViewTable,
  OssieRelationshipViewTable — Calcite Table / Schema
  implementations.
• OssieAutoJoinRule — Calcite RelRule that walks the semantic
  model's relationships block and injects join predicates at plan
  time so queries can name multiple datasets in FROM without
  writing JOIN ... ON ....
• OssieShelfSqlTranslator — compiles OssieQuery → SQL by walking
  the bi.saiku.ossie.model.SemanticModel DTO tree directly.

Three ways to query

1. Shelf-state builder (engine.execute(OssieQuery)) — for
   agents, workbench UIs, and code that wants a typed API.
2. Raw SQL passthrough (engine.executeSql(sql)) — for BI
   query editors, LLMs that write joins directly, or anything that
   speaks SQL.
3. JDBC connection (engine.openConnection()) — the same
   Calcite JDBC connection under the hood; any tool that speaks
   JDBC gets a semantic-aware surface.

Multi-module layout

ossie-core and ossie-sql now live under a parent pom that
manages shared versions (Jackson, Calcite, JUnit, Guava, etc.).
Each child module has its own release lifecycle but ships in the
same repo + one CI matrix.

Tests

10 new tests in OssieEngineTest — real end-to-end against an
in-process H2 warehouse seeded with an orders fixture:

• Simple aggregation (no dimensions) executes and returns typed
  numbers
• Cross-dataset join (customers × orders) auto-resolves via the
  relationships block — no manual JOIN in the request or the SQL
• Ratio metric composes SUM(...) / NULLIF(COUNT(*), 0) inline
• Filter IN excludes rows correctly
• SQL preview (compile()) works without hitting the warehouse
• Raw SQL passthrough via executeSql() uses the same schema
• openConnection() yields a usable JDBC handle
• Unknown metric names throw with the semantic model name in the
  message
• Result carries the generated SQL string
• Column descriptors carry the right type ("dimension" vs
  "metric")

Full suite (34 tests): 34/34 green.

README

New top-level structure listing both modules. Detailed ossie-sql
section with:

• Quick start dependency block
• End-to-end example (build engine, build query, execute, print
  records)
• SQL preview example
• Raw SQL over the semantic model example
• What's supported list
• Warehouse compatibility list